---
license: mit
library_name: onnx
tags:
  - text-to-speech
  - indic
  - onnx
  - vocos
  - f5-tts
  - indian-languages
pipeline_tag: text-to-speech
base_model: ai4bharat/IndicF5
---

# IndicF5 ONNX

ONNX exports of [ai4bharat/IndicF5](https://huggingface.co/ai4bharat/IndicF5) (DiT denoiser + Vocos vocoder) for local/CPU inference with ONNX Runtime.

## Files

| File | Notes |
|------|--------|
| `indicf5_dit_fp32_static.onnx` | DiT, fp32, static shapes (~1.3 GB) |
| `indicf5_vocos_fp32.onnx` | Vocos, fp32, dynamic frames |
| `indicf5_vocos_fp32_static.onnx` | Vocos, fp32, static |
| `indicf5_vocos_fp16.onnx` | Vocos, fp16 |
| `indicf5_vocos_int8.onnx` | Vocos, int8 |

Sample WAVs and mel inputs are under `samples/` (Hindi + Tamil).

## Notes

- Base model: IndicF5 (~0.35B F5-TTS-style TTS for Indian languages).
- Typical path used here: PyTorch/ONNX DiT mel → ONNX Vocos decode.
- Not realtime on Apple M1 CPU in our benches (DiT dominates).

## Attribution

Derived from [ai4bharat/IndicF5](https://huggingface.co/ai4bharat/IndicF5). Follow the base model license/terms for redistribution and use.